Tech is the future but it won't look like the past
Our tech future is a high-capex, industrial, one, not a software-driven one
Welcome to the latest edition of Buy the Rumor; Sell the News. In today’s post, I reconcile my realist views about technology and investment returns with my overall optimism about the arc of technological progress.
If you like what you read here, consider subscribing, and if you’re already subscribed, consider upgrading to a paid subscription. If you want to connect with me directly, my contact information is at the end of this post.
The paradox
At first glance, my views might seem paradoxical. I’ll tell you that humanoid robots will mostly flop, that hyperscaler returns on invested capital will grind lower under enormous capex burdens, that startups built on top of foundation models will be crushed by commoditization. I’ll also say that venture capital’s power law is under assault because the new frontier is so capital-intense that it simply can’t sustain asymmetric payoffs like it once did.
But then, in the next breath, I’ll insist that technology is still the most powerful force shaping our collective future, that there is no serious ceiling on AI capabilities, on bioengineering, on new materials, on synthetic energy systems. That even the geopolitical game is now fundamentally a technological arms race.
How do these two lines of thought coexist? Isn’t it contradictory to be both deeply bullish on tech’s transformative arc and sharply skeptical of near-term business models, margins, and investment returns?
Not at all. The deeper truth, and the alpha insight for both builders and investors, lies precisely in reconciling them.
The inevitability of progress (on long timeframes)
Start with the biggest picture.
Technological advance is not slowing. In fact, on almost any meaningful scale, it’s accelerating. AI is already fundamentally different from three years ago, and the slope of compute-driven discovery is steepening. Meanwhile, bioinformatics, energy systems, advanced manufacturing, even our understanding of cognition itself are all being pulled forward by tools that didn’t exist five years ago.
We’re rewriting the substrate of civilization. In twenty years, a typical day on Earth will involve more ambient intelligence, more self-optimizing systems, more engineered biology, more synthetic energy than we can fully grasp today. That is not hype. That is simply the compounded trajectory of discovery, tool-building, and market selection.
But progress ≠ margins or venture-like returns
Here’s where people get tripped up.
It does not follow that because technology is unstoppable, investing in or building the wrong companies will be lucrative. Nor does it mean the returns to capital will look anything like the last twenty years of software-heavy, capex-light scaling.
In fact, the very nature of this next phase of technological deployment virtually ensures it won’t.
AI at the frontier is now infrastructure. It’s national capability, not simply commercial opportunity. That means it’s going to be owned or heavily shaped by sovereign interests, via regulation, direct capital injections, defense programs, or just brute sovereign power. The profit pools will exist, but they won’t be clean software take-rates.
The capex has become massive. Training frontier models, building advanced fabs, deploying high-density energy systems, or rolling out biomanufacturing platforms isn’t cheap. The days of two kids in a garage spinning up billion-dollar value on AWS with minimal fixed costs are gone for these new verticals.
Returns on invested capital will compress. That’s the brutal math of capital intensity. When the moat becomes “spend $20 billion on plants, GPUs, power contracts, and fluid cooling,” your IRRs inevitably look more like regulated utilities or old industrials.
Why most startups will get slaughtered
This is why I’m perfectly comfortable saying most foundation model-based startups are doomed.
They’re operating on an outdated power law playbook: low capital, massive TAM, thin teams. They imagine that if they ride the wave of model progress, some thin layer of IP or brand or UX differentiation will carry them to multi-billion valuations. For the most part, it won’t.
Almost everything they’re building is easily replicable by the next open checkpoint or by an enterprise deal with OpenAI, Anthropic, or whoever’s running the best GPU clusters that month. Their venture investors keep hoping to replay the 2012-2020 SaaS game. Meanwhile the economics are screaming: this is different.
So why still be bullish?
Because history isn’t a story about VC multiples. It’s a story about the evolution of intelligence and control over matter. That’s continuing, with or without clean startup exits.
And, paradoxically, if you have the right frame, there are actually better opportunities now, even if they don’t look like classic venture:
Owning the inputs. Power, land, water, industrial real estate, specialized cooling all suddenly matter. The new bottlenecks are starkly physical. Some of the highest IRRs of this era may come from dull-sounding places like transmission, grid-scale battery arbitrage, or water rights leveraged for data center expansions.
Building to acquisition. As hyperscalers consolidate, many small advanced tech companies will find their best exits in M&A, not IPOs. That still creates wealth, just through a different vector.
Sovereign-aligned scale. If you can operate in the slipstream of national security budgets or climate transition mandates, you unlock non-dilutive capital and a strategic moat. That’s far removed from consumer SaaS or even standard enterprise software.
Specialized operating expertise. The most scarce resource is no longer app developers or even prompt engineers. It’s people who can navigate giga-scale construction, negotiate power off-take, or secure export-controlled lithography tooling.
Reconciliation of the paradox
So yes, I believe humanoid robots will mostly flop in the first wave. Their failure modes are obvious: unstable edge cases, social acceptability, insurance and liability regimes, absurd maintenance costs. But some version will still iterate forward. Twenty years from now, we’ll see robots handling certain industrial or logistics tasks because we’ll have figured out how to do it reliably and economically.
Yes, I think most venture capital chasing large language model wrappers will be incinerated. But the overall intelligence capacity of our civilization is still increasing, creating profound downstream effects in science, logistics, policy, and even existential risk management.
Yes, hyperscaler returns will trend toward regulated utility profiles. That’s not sexy. It’s also not the end of progress. It’s a maturation phase where the infrastructure of intelligence becomes as essential, and as politically entangled, as power grids or telecom networks.
What it all means
It’s only a contradiction if you mistake technological advancement for venture returns, or startup stories, or even private equity-style cash flows. The real story is bigger and stranger:
The future is coming. The arc of intelligence, automation, and engineered matter is unstoppable.
But it will arrive through heavy industry, sovereign programs, and capital structures that look very different from the last cycle.
To thrive, you need to shed the narratives of the 2010s and understand the new game: who owns the pipes, the land, the watts, the political relationships, the sovereign guarantees.
That’s not dark. It’s just more textured than fairy tales of endless software margins. In the long run, it’s arguably even more exciting, because it means we’re about to rebuild civilization’s entire substrate. And we get to decide, through the capital we deploy and the systems we build, what kind of world that becomes.
Coda
If you enjoy this newsletter, consider sharing it with a colleague.
Most posts are public. Some are paywalled.
I’m always happy to receive comments, questions, and pushback. If you want to connect with me directly, you can: